Update app.py

app.py

@@ -18,51 +18,33 @@ from huggingface_hub import InferenceClient, login
 
 class BasicAgent:
     def __init__(self):
-        # Initialize with GAIA-proven Qwen model
+        login(token=os.environ["HF_TOKEN"])  # Required for gated access
+        
         self.client = InferenceClient(
-            model="Qwen/Qwen2-72B-Instruct",
+            model="Qwen/Qwen2-7B-Instruct",
             token=os.environ["HF_TOKEN"],
             timeout=120
         )
         
-        # Verify model access
-        test_response = self._call_model("2+2=")
-        if not test_response.startswith("4"):
-            raise RuntimeError("Model initialization failed")
-
-    def _call_model(self, question: str) -> str:
-        """Core model call with GAIA-optimized prompt"""
-        prompt = f"""<|im_start|>system
-Answer with ONLY the exact value requested, no explanations. Follow GAIA format strictly.<|im_end|>
+    def __call__(self, question: str) -> str:
+        try:
+            prompt = f"""<|im_start|>system
+Answer with ONLY the exact value requested.<|im_end|>
 <|im_start|>user
 {question}<|im_end|>
 <|im_start|>assistant
 """
-        return self.client.text_generation(
-            prompt=prompt,
-            temperature=0.05,
-            max_new_tokens=100,
-            stop_sequences=["<|im_end|>"],
-            repetition_penalty=1.1
-        )
-
-    def __call__(self, question: str) -> str:
-        try:
-            # Get raw model response
-            raw_response = self._call_model(question)
-            
-            # Strict GAIA-compliant extraction
-            answer = raw_response.split("<|im_start|>assistant")[-1]
-            answer = answer.split("<|im_end|>")[0].strip()
-            
-            # Normalization for exact matching
-            return re.sub(r'[^a-zA-Z0-9, ]', '', answer).lower()
-        except Exception as e:
-            print(f"Error: {str(e)}")
+            response = self.client.text_generation(
+                prompt=prompt,
+                temperature=0.01,
+                max_new_tokens=50,
+                stop_sequences=["<|im_end|>"]
+            )
+            return response.split("<|im_start|>assistant")[-1].split("<|im_end|>")[0].strip()
+        except Exception:
             return ""
 
 
-
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
     Fetches all questions, runs the BasicAgent on them, submits all answers,